ai agents
The Number That Tells You Whether Your AI Agent Is Actually Getting Better
Most teams can say their AI agent looks good today, not whether it beat last Tuesday. How a fixed test set, guardrail metrics and a release gate fix that.
Field notes on agentic commerce and AI agents
When the agent is the buyer, what does your product have to become?
Saturnia Labs writes about what changes when AI agents act for people: catalogs agents can read, payments under a mandate, verification and disputes, the brand work that moves off the homepage, and how to build and evaluate the agents themselves. Written by the team that makes stores agent-ready.
Latest
All posts
Subscribe via RSS- 22
The Number That Tells You Whether Your AI Agent Is Actually Getting Better
Most teams can say their AI agent looks good today, not whether it beat last Tuesday. How a fixed test set, guardrail metrics and a release gate fix that.
- 21
The model is the cheapest part of your product
Any competitor can rent your model by lunchtime. The layer above it, task contract, context, permissions, routing, memory, evals, economics, is the moat.
- 20
I promoted myself out of the only work that didn't get cheaper
Vibe coding your MVP buys speed and hidden coupling. Building got cheap, architecture did not: five things to write down before you open the agent.
- 19
You bought six AI tools and the work multiplied
Six competent AI tools, arranged badly, create supervision work nobody tracks. Why the METR study collapsed, and a five-day log that shows where hours go.
- 18
The Agent That Waits Is a Chatbot
A chatbot returns an answer; an agent changes a state. Five verbs (wake, observe, decide, act, report) and six Tuesday-morning questions for a demo.
- 17
Verification Is the Product: Fraud in Agentic Commerce
When an AI agent buys on your behalf, the human fraud checks leave with the human. Why verifying buyer, agent, merchant and mandate becomes the product.
- 16
The Agent Can Pay, But Who Gave It Permission?
In agentic payments the mandate replaces the checkout click. What it must record (scope, threshold, policy, escalation), and why revocation is a feature.
- 15
Your Catalog Becomes Your Sales Team
AI shopping agents never load your homepage. They read product data, check freshness, and choose. Why catalog quality now decides if agents surface you.
- 14
Shopping Moves From Search to Instruction
The unit of commerce was a page, then a cart, now a task. How instruction-led shopping changes intent capture, personalization and the merchant's job.
- 13
Brand Strategy After the Homepage Loses Its Monopoly
Shopping agents never see your hero image. They read product data, policies and fulfillment history. Where brand strategy goes when an agent makes the cut.
- 12
Trust Becomes the Payment Layer
Agents can already buy. The slower problem is whether people will let them. Why permission grows one grant at a time and consent is the payment layer.
- 11
The Agent Becomes the Buyer Interface
Journeys now start inside a chat with an agent. How the funnel compresses into a brief, why merchants lose the first impression, and what earns a slot.
- 10
The New Shelf Space Is Structured Data
Agents rank products on structured fields, not marketing copy. Why identifiers, spec fields and feed latency now decide whether a product is on the shelf.
- 09
Are You Ready for Agentic Payments? A 25-Point Audit
A 25-point audit for PSPs and merchants before agents initiate payments: 15 system checks on tokens, SCA and disputes, 10 behavior checks on user control.
- 08
Disputes Will Make or Break Agentic Payments
When a user says this wasn't me about an agent's purchase, the provider must prove mandate, decision and authentication. A dispute-first architecture.
- 07
Trusted Agent or Malicious Bot? A Merchant Framework for Telling the Difference
Bots are 51% of web traffic, but a customer's shopping agent is demand you cannot block. A four-layer merchant framework for telling them apart.
- 06
Standing Intents: The Agent That Keeps Shopping When You've Already Closed the Tab
Standing intents turn a payment into a policy: subscriptions, replenishment and price-triggered buying, what they need from infrastructure and merchants.
- 05
One Cart, Many Merchants: The Drop-Ship Logic That Makes Agents Worth Using
People shop by project, checkout runs by merchant. How an agent holds a multi-merchant bundle together, what the review screen must show, and what breaks.
- 04
Mandates as Product: Designing Consent for Agentic Payments
Agentic payments fail first on one question: what did the user allow? How to design the mandate as a product: budget, scope, time, escalation, revocation.
- 03
Agentic Payments 101: From "I Want This" to a Settled Transaction
What agentic payments are, the four artifacts that make delegation enforceable, three transaction patterns, and the eight steps from intent to settlement.
- 02
Founder Bias in User Research: How to Hear the Truth Without Triggering Defense Mode
Users say it looks good, the funnel stays flat. A No-Defense Pact, behavior-first prompts and a facilitator-observer setup turn polite sessions into proof.
- 01
The Third-Party Truth Audit: A 10-Day UX Sprint That Finds Revenue-Blocking Bottlenecks
Traffic and signups but flat activation? A ten day, neutral facilitator UX sprint that tests your money paths and delivers a bottleneck backlog, not a deck
Is your shop ready for shopping agents?
Find out what an AI agent sees when it reads your store.
Run the free readiness scan on any store URL, or book a 30 minute readiness audit: a score for your catalogue and checkout, and a live agent demo against your own products.